Abstract
While evolutionary computation and evolutionary robotics take inspiration from nature, they have long focused mainly on problems of performance optimization. Yet evolution in nature can be interpreted as more nuanced than a process of simple optimization. In particular, natural evolution is a divergent search that optimizes locally within each niche as it simultaneously diversifies. This tendency to discover both quality and diversity at the same time differs from many of the conventional algorithms of machine learning, and also thereby suggests a different foundation for inferring the approach of greatest potential for evolutionary algorithms. In fact, several recent evolutionary algorithms called quality diversity (QD) algorithms(e.g. novelty search with local competition and MAP-Elites) have drawn inspiration from this more nuanced view, aiming to fill a space of possibilities with the best possible example of each type of achievable behavior. The result is a new class of algorithms that return an archive of diverse, high-quality behaviors in a single run. The aim in this paper is to study the application of QD algorithms in challenging environments (in particular complex mazes) to establish their best practices for ambitious domains in the future. In addition to providing insight into cases when QD succeeds and fails, a new approach is investigated that hybridizes multiple views of behaviors (called behavior characterizations) in the same run, which succeeds in overcoming some of the challenges associated with searching for QD with respect to a behavior characterization that is not necessarily sufficient for generating both quality and diversity at the same time.
Highlights
The products of nature have long served as inspiration for the investigation and practice of evolutionary algorithms and evolutionary robotics (Cliff et al, 1993; Nolfi and Floreano, 2000; Stanley, 2011)
Its main difference from quality diversity (QD) is that MMFO traditionally focuses on genetic diversity and tends to apply only to simple phenotypes, such as mathematical functions, where the genotype and phenotype are in effect the same (Mahfoud, 1995); QD reflects a later shift in interest toward behavioral diversity and is often applied in domains such as evolutionary robotics where the relationship between genome and behavior is complex
In all of the figures presented treatments are color coded according to which behavior characterization (BC) drives search: DirectionBC is drawn in blue, EndpointBC in yellow, multi-BC in green, and Fitness in gray
Summary
The products of nature have long served as inspiration for the investigation and practice of evolutionary algorithms and evolutionary robotics (Cliff et al, 1993; Nolfi and Floreano, 2000; Stanley, 2011). The ability of such algorithms to match the complexity and sophistication of nature has frustratingly lagged, as researchers in the fields often observe (Stanley and Miikkulainen, 2003; Doncieux et al, 2015). The main limitation of genetic diversity is that it is susceptible to genetic aliasing, which means that genomes that are different may behave Such aliasing, which is amplified especially in the presence of indirect genotype to phenotype mappings (Hornby and Pollack, 2001, 2002; Bongard, 2002; Stanley and Miikkulainen, 2003; Stanley, 2007), is counterproductive to find various behaviors, as shown empirically by Trujillo et al (2011)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
